INTRODUCTION TO
MACHINE
[Link]@[Link]
DSEFTCJUL18029
LEARNING
Dr. Narayana
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Human Body Temperature Distribution
According to Their Emotional State
[Link]@[Link]
DSEFTCJUL18029
Source: [Link]
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Human emotions and
postures
[Link]@[Link]
DSEFTCJUL18029
Source: “Recognizing Emotions Expressed by Body Pose: a Biologically Inspired Neural Model” by Schindler et al,
Neural Networks, 2008, [Link]
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Threat Perception in
Real Time Security Systems
■Can we recognize emotions
– from posture and thermographs?
– of an individual in a group or crowd?
■Can we link recognized emotions to predict
[Link]@[Link]
DSEFTCJUL18029
possible threat from an individual in a crowd
monitored using thermographs and gait
analysis from video cameras?
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
What resolution? Video camera
Correlate and Classify Acquired Images
Infra Into Predefined Set
red camera of devices
Storage
Emotions Based on Temperature Distribution And Postures
Data transfer speeds
Programmable chips Data?
Good accuracy
Algorithm
Less uncertainty
Identify Emotions from
Work for group
[Link]@[Link]
DSEFTCJUL18029
Body Temperature
of humans? Distributions
and Posture of humans
Programmable?
Execution time?
Is it scalable?
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Threat Perception in
Real Time Security Systems
■ Desired features of an algorithm
– Programmable/ Automate
– Realistic computation time
– Accurate and Precise, independent of data
accuracy/precision
■ Considerations for the data
[Link]@[Link]
DSEFTCJUL18029
– Required accuracy and precision in thermograph and
video resolution, magnification
■ How can we automate/program this task?
Possibly embed this into camera chip and link it to
the security system?
See: Multi-sensor Surveillance Systems: The Fusion Perspective, edited by Gian Luca Foresti, Carlo S. Regazzoni, Pramod
K. Varshney, Google books URL: [Link]
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Outline
■ What is Machine Learning?
■ What machine learning can do?
■ Machine learning models/algorithms
[Link]@[Link]
■ Design of Machine Learning Study
DSEFTCJUL18029
■ Goals and Objectives of This Module
■ Summary and Comments
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
What is “Machine Learning”?
■ Algorithms and techniques used for data analytics
■ Studies how to automatically learn to make accurate
predictions based on past observations
■ Machine learning is programming computers to
optimize a performance criterion by tuning set of
[Link]@[Link]
DSEFTCJUL18029 parameters . These tuned programs then perform
same task on unseen data.
New
Examples
Machine
Training Classification
Learning
Examples Rule
Algorithm
Predicted Class
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
[Link]@[Link]
DSEFTCJUL18029
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Machine Learning is
used when…
■ Human expertise does not exist
– Navigating on Mars
■ Humans are unable to explain their expertise
[Link]@[Link]
– speech recognition
DSEFTCJUL18029
– mine detection
■ Solution changes or evolves in time
– routing on a computer network
■ Solution needs to be adapted to particular cases
– user biometrics, virtual agent based solutions
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Applications
■ Retail: Market basket analysis, Customer relationship
management (CRM)
■ Finance: Credit scoring, fraud detection
■ Manufacturing: Optimization, troubleshooting
■ Medicine: Medical diagnosis
[Link]@[Link]
DSEFTCJUL18029
■ Telecommunications: Quality of service optimization
■ Bioinformatics: Motifs, alignment
■ Web mining: Search engines
■ And many more….
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Algorithms
■ The success of machine
learning system also Machine Learning
depends on the Algorithms
algorithms.
■ The algorithms control
Supervised Learning
[Link]@[Link]
the search to find and
DSEFTCJUL18029
build the knowledge
structures. Unsupervised
Learning
■ The learning algorithms
should extract useful Reinforcement
information from training Learning
examples.
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Supervised Learning
■ Learning from examples
■ We are given attributes, 𝑿 and Training Feature
targets 𝒚 Data Extraction
■ knowledgeable external 𝑿
supervisor 𝒚
– Regression
[Link]@[Link]
DSEFTCJUL18029 ML Model
– Classification
𝒘
– Decision trees
– Random forest ML Algorithm
Performance
metric
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Regression: Examples
Reading your mind Predicting stock prices
■ Happiness state is related to ■ Depends on
brain region intensities – Recent stock prices
– News Events
[Link]@[Link]
DSEFTCJUL18029
– Related commodities
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Classification: Examples
Credit scoring Outlook of the day and Weather
■ Differentiating between low- derivatives
risk and high-risk customers
from their income and
savings
[Link]@[Link]
DSEFTCJUL18029
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Classification: Applications
Also known as Pattern recognition
■ Face recognition: Pose, lighting, occlusion (glasses,
beard), make-up, hair style
■ Character recognition: Different handwriting styles.
■ Speech recognition: Temporal dependency.
– Use of a dictionary or the syntax of the language.
[Link]@[Link]
DSEFTCJUL18029
– Sensor fusion: Combine multiple modalities; eg,
visual (lip image) and acoustic for speech
■ Medical diagnosis: From symptoms to illnesses
■ Web Advertising: Predict if a user clicks on an ad
on the Internet.
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Unsupervised Learning
■ Learning from
examples Training Feature
■ We are given only Data Extraction
Intelligence:
attributes, 𝑿, and no 𝑿 Segmentation,
targets Pattern, Cluster
[Link]@[Link]
– Clustering
DSEFTCJUL18029 ML Model
– Finding association (in 𝒘
features)
– Image compression
– Probability distribution ML Algorithm
estimation
– Dimension reduction
Performance
metric
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Learning Associations
■ Basket analysis:
P (Y | X ) probability that somebody who buys X also
buys Y where X and Y are products/services.
Example: P ( chips | beer ) = 0.7
[Link]@[Link]
DSEFTCJUL18029 Market-Basket transactions
TID Items
1 Bread, Milk
2 Bread, Diaper, Beer, Eggs
3 Milk, Diaper, Beer, Coke
4 Bread, Milk, Diaper, Beer
5 Bread, Milk, Diaper, Coke
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Object recognition
Image Object recognition Blind source separation
■ Recognize objects in the ■ Recognize source/s in a
image mixed music signal
[Link]@[Link]
DSEFTCJUL18029
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Reinforcement learning
■ Mimics intelligent system
■ Observers interaction of
environment and system
actions
■ Optimize goal/rewards
[Link]@[Link]
■ Continuous, self-learning
DSEFTCJUL18029
Image Source: [Link]
■ It is not a method but a • Applications
process as a whole to build – Decision making
knowledge
– Robot, Chess machine
■ Corrective action even if – Optimal control theory
system sees a new situation
– Stochastic approximations
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Machine Learning and
Traditional Statistics
Machine Learning Traditional Statistics
■ Emphasize predictions, ■ Emphasizes super-
usually no super- population inference
population model specified ■ Focuses on a-priori
[Link]@[Link]
DSEFTCJUL18029 ■ Evaluates results via hypotheses
prediction performance ■ Simpler models preferred
over complex ones
■ Concern for overfitting but (parsimony), even if the
not model complexity per more complex models
se perform slightly better
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Machine Learning and
Traditional Statistics Continued…
Machine Learning Traditional Statistics
■ Emphasis on performance ■ Emphasis on parameter
interpretability
■ Generalizability is obtained
through performance on ■ Statistical modelling or
novel datasets sampling assumptions
connects data to a
[Link]@[Link]
DSEFTCJUL18029
■ Concern over performance population of interest
and robustness ■ Concern over assumptions
and robustness
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Machine learning design
study: Data science process
[Link]@[Link]
DSEFTCJUL18029
Source: “Doing data science” by Schutt, R and O’Neil, C
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Machine learning design
study: Data scientist role
[Link]@[Link]
DSEFTCJUL18029
Source: “Doing data science” by Schutt, R and O’Neil, C
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Machine Learning Algorithms
Supervised Learning Unsupervised Learning
■ Regression ■Clustering &
Dimensionality
Continuous
– Ordinary Least Squares
Reduction
■ Decision Trees – K-means
■ Random Forests – SVD
[Link]@[Link]
– PCA
DSEFTCJUL18029
■ Classification
Categorical
– K-Nearest Neighbour
(KNN) ■Association Analysis
– Logistic Regression – Apriori
– Trees – Market basket analysis
– Support Vector Machine
(SVM)
– Naïve-Bayes
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Course philosophy: Always
use case study and ..
Core
Visual Algorithm
Concept
[Link]@[Link]
DSEFTCJUL18029
Advanced
Practical Implement
Concepts
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Resources: Datasets
■ UCI Repository:
[Link]
■ UCI KDD Archive:
[Link]
[Link]@[Link]
DSEFTCJUL18029
■ Statlib: [Link]
■ Delve: [Link]
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Summary
■ We had a broad overview of
– machine learning philosophy
– Different algorithms
[Link]@[Link]
DSEFTCJUL18029 – Application fields of machine learning
– Examples
■ Provided links to datasets
■ We also introduced
– Course content
– recommended approach to master machine learning
techniques
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.
Thank you!
[Link]@[Link]
DSEFTCJUL18029
This file is meant for personal use by [Link]@[Link] only.
Sharing or publishing the contents in part or full is liable for legal action.